Plasma Proteomics Maps Aging Risk by Organ and Sex

Plasma proteomics is making biological-age clocks more specific. Instead of returning one composite number that says you are biologically older or younger than your birth certificate, newer models try to identify whether the strongest aging-linked signal comes from the heart, kidneys, brain, immune system, or another organ-related domain. Large studies also show that protein patterns differ by sex and can predict risk across hundreds of diseases.

That is useful progress in risk mapping. It is not proof that a high "heart age" causes heart disease, that a protein drives aging, or that lowering a score with a supplement will improve health. For now, these measurements are research-grade maps: detailed, potentially informative, and far ahead of the evidence needed to prescribe an intervention.

From one biological age to several organ-linked signals

The clearest example comes from Oh et al.'s 2023 Nature study. The researchers used SomaScan measurements of plasma proteins to build 11 clocks based on proteins classified as enriched in particular organs. They began with 4,979 analytes, identified 893 organ-enriched proteins through GTEx tissue-enrichment rules, retained 856 after quality control, and trained the clocks in 1,398 healthy Knight-ADRC participants.

For each clock, the researchers calculated an organ "age gap": predicted organ age minus chronological age. A positive gap meant that the protein pattern resembled that of an older person. The labels require some caution. A plasma protein assigned to the kidney by tissue-enrichment rules is not a direct measure of kidney tissue, much less a biopsy. "Kidney age" is a statistical score built from plasma abundance and an organ-enrichment framework.

The clocks predicted chronological age across several external cohorts, including Covance, LonGenity, SAMS, and Stanford-ADRC. More interestingly, the organ gaps did not move in lockstep. Their mean inter-organ correlation was only r=0.29. In the study population, 23% of participants qualified as extreme agers in at least one organ; 18.4% had a single-organ extreme ageotype, while 1.7% had multi-organ extreme aging. Those patterns support the idea that circulating proteins can capture partly separable age-related variation rather than one universal decline score.

The disease associations were most persuasive when the score and disease occupied the same broad physiological territory. Kidney age gaps were associated with hypertension and diabetes, while heart age gaps were associated with atrial fibrillation and previous myocardial infarction. Depending on the condition, the estimated acceleration ranged from about 1.0 to 2.8 years. These are observational associations, not evidence that the measured proteins caused the disease or that changing the score would change the outcome.

Prediction can be clinically interesting without being a treatment target

Oh et al. also tested prospective outcomes. Among 812 LonGenity participants without heart failure at baseline, heart age acceleration predicted incident heart failure over 15 years: hazard ratio 2.37, based on 26 events. Ten of the 11 organ gaps also associated with all-cause mortality, largely in the same cohort. The result is notable, but the small number of heart-failure events makes the estimate imprecise in practical terms, and the study design cannot fully separate upstream aging biology from early, undiagnosed disease.

That distinction matters because plasma proteins already predict prognosis in people who have established disease. In a 2022 heart-failure study by Gui et al., a plasma proteomic profile predicted survival in patients with heart failure with reduced ejection fraction. That finding supports proteomics as a prognostic tool. It also offers a competing explanation for some aging-associated signatures: proteins can reflect tissue injury, congestion, inflammation, medication exposure, or illness severity. A score may forecast a bad outcome because it detects disease burden, not because it has measured an independent biological aging process.

Methodological choices add another layer of uncertainty. Drake et al. showed that high-dimensional plasma-protein associations with all-cause mortality can change with modeling decisions, covariates, multicollinearity, and multiple-testing strategy. Mortality prediction is valuable, but it does not automatically identify the machinery of aging.

A population atlas reveals scale—and non-specificity

Deng et al.'s 2025 Cell study expanded the field from organ clocks to a disease atlas. The investigators measured 2,920 plasma proteins in 53,026 UK Biobank participants and followed them for a median of 14.8 years. They reported 168,100 protein-disease associations and 554,488 protein-trait associations. Protein panels achieved an area under the receiver-operating-characteristic curve above 0.80 for 183 diseases.

An AUC above 0.80 can indicate useful discrimination, but it does not by itself establish clinical usefulness. A model can distinguish cases from non-cases yet fail to improve decisions over age, medical history, standard laboratory values, genetics, or existing risk scores. The published record available here does not provide enough disease-by-disease comparison to verify broad superiority over standard care.

The atlas also complicates tidy organ stories. More than 650 proteins were associated with at least 50 diseases. That pleiotropy may reflect shared inflammation, metabolic stress, tissue repair, or correlated illness. It means a protein can be a strong predictor while offering weak evidence for a single-organ mechanism. The study's genetic integration prioritized 474 candidate causal proteins, 37 drug-repurposing opportunities, and 26 targets with favorable safety profiles. Those are sensible leads for experimental work, not proof that altering any of those proteins will slow aging or prevent disease in people.

Sex-specific trajectories make a universal score less convincing

Aging maps also need sex-specific interpretation. In a 2025 Communications Medicine analysis, Jin et al. examined 2,923 Olink proteins in 53,013 UK Biobank participants. Female protein trajectories showed more nonlinear age patterns and more proteins with accelerating aging rates. Yet the researchers found many more proteins jointly associated with aging and mortality in males: 172 in males versus one in females, including male signals such as GDF15 and IL1R2.

This does not mean that women have one mortality-relevant protein or that men age more quickly. It means the associations, as modeled in this cohort, differed sharply by sex. Menopause, body composition, medications, comorbidities, and lifestyle patterns could contribute to those differences. Replication beyond a UK Biobank-like population and careful adjustment for those factors will determine how much of the pattern reflects sex-specific biology.

The brain signal is promising but less settled

Brain aging is where the limits of transportability become especially visible. In Oh et al., a first-generation plasma brain-age gap associated with Alzheimer's disease in held-out Knight-ADRC cases but did not replicate in Stanford-ADRC. The authors obtained stronger findings only after selecting features around cognition. That is a reasonable modeling response, but it means the original score was not a stable, externally replicated measure of brain age across those cohorts.

Other studies still support the broader idea that plasma proteins can map brain-related risk. Liu et al. identified plasma biomarkers and dynamic patterns linked to brain aging. In two cohorts, Huang et al. found proteins associated with Alzheimer's diagnosis and blood amyloid, tau, neurodegeneration, and inflammation biomarkers; 20 of 35 proteins associated with Alzheimer's disease replicated in an independent cohort. These studies show biologically coherent associations. They do not establish that the proteins are direct levers of brain aging.

Plasma proteomics can sort people into more detailed risk patterns than a single biological-age score. It has not yet shown which score-guided action improves a patient outcome.

What would make these maps actionable?

First, researchers need external replication across ancestries, health systems, and laboratory settings. Assay dependence is a practical obstacle: Oh et al. used SomaScan, while the large UK Biobank studies used Olink Explore panels. These platforms measure overlapping but nonidentical protein spaces, and no cited study establishes universal calibration or a common clinical cut point.

Second, scores need prospective tests against ordinary care. A useful trial would ask whether knowing a proteomic risk score changes screening, treatment, or prevention decisions and whether those changes improve hard outcomes. It would also need to show net benefit, not merely a more impressive graph or a lower biomarker score.

For biohackers, the sensible stance is restrained curiosity. A plasma proteomic report may eventually help clinicians identify risk patterns or help researchers select trial participants. Today, no cited study provides validated thresholds for changing supplements, prescription drugs, screening intensity, or lifestyle based on an organ-age gap. Eat well, exercise, avoid smoking, manage blood pressure and lipids, and use established preventive care. The protein map may become more useful; those foundations already are.